Enhanced Intraoperative Visualization for Brain Surgery: A

Transcription

Enhanced Intraoperative Visualization for Brain Surgery: A
Enhanced Intraoperative Visualization for Brain Surgery:
A Prototypic Simulated Scenario
Z. Salah1,2, B. Preim1, A. Samii3, R. Fahlbusch3, and G. Rose2
¹ University of Magdeburg, Institute for Simulation and Graphics, Magdeburg, Germany
² University of Magdeburg, Institute for Electronics, Signal Processing and Communications, Magdeburg, Germany
3
International Neuroscience Institute, Hannover, Germany
Contact: [email protected]
Abstract:
Due to the required high precision and reliability, image guidance is increasingly being adapted in neurosurgery, where
correspondences between the patient and scanned data are viewed on multiple displays. In this paper, we present a prototype for an augmented reality system that provides enhanced intraoperative visualization within the optical surgery
microscope image. For this purpose, relevant targeted and risk structures are extracted from registered patient data and
overlaid on the real scene. Similarly, supportive information of underlying anatomy can be superimposed, which includes, amongst others, shape and depth information of tumors, and interactively selected tomographical slices. The
overall advantage is a more ergonomic hand-eye coordination, a more accurate delineation of targeted structures and
better protection of risk anatomy.
Keywords: Medical Augmented Reality, Intraoperative Visualization, Medical Navigation
1
Problem
Medical navigation systems help physicians to establish correspondences between locations in an acquired patient dataset and the patient's physical body during navigated surgeries. This is highly advantageous in neurosurgery, where high
precision is demanded. In a microscope-assisted surgery, however, this requires the surgeon to switch between the microscope and wall-mounted or computer displays. By incorporating augmented reality, it becomes possible to view patient data directly in the 3D context of the patient himself. This may provide an in-place insight into the patient body,
which, for example, would help defining the entry point and the trajectory of a biopsy needle [7]. Some systems have
been introduced that provide enhanced endoscopic views that are paired with synthesized virtual renderings generated
from the same view [4, 6]. Other systems tried to modify the design of operating binoculars [2] and microscopes [3] to
allow for data augmentation.
In this paper, we present a prototype for enhancing intraoperative visualization within the operating microscope, by
augmenting the video stream with relevant patient data from different diagnostic and intraoperative imaging modalities.
The advantage is a better localization of targeted lesions (e.g. tumor) and the protection of hidden risk structures like
blood vessels and neural connections. The prototype includes the different components of a navigated procedure: tracking, calibration, registration, and rendering, which are described in the following section.
2
Methods
For our prototype, we implement a marker-based optical tracking system and use a tablet PC with a high resolution
built-in camera to simulate the surgical microscope. A phantom model of the head simulates the patient. In the context
of this paper, we refer to the tablet PC camera as the video camera and to the camera of the tracking system as the
tracker camera. At a first step, the system cameras are calibrated to compute their intrinsic geometrical parameters.
This is performed by capturing several photos of a special checkerboard pattern, with the tracker camera, from different
Proceedings curac2010@MEDICA
125
locations and orientations. Afterwards, MATLAB’s Camera Calibration Toolbox1 is used to calculate the focal length,
principal point, and skew and distortion coefficients of the camera. The same process is performed for the video camera.
The intrinsic calibration procedure is performed only once and the camera parameters are stored.
A tracking server continuously captures a video stream with the tracker camera (Logitech QuickCam Pro 9000; Resolution: 1280×720 pixel; 30 fps). At each frame, the system searches for predefined markers and, for each detected marker,
computes the 3D position and orientation in the camera coordinate system. Pose information of all detected markers are
broadcasted to the tablet PC (and other devices, if necessary) over a WLAN connection.
Figure 1: Calibration of the video camera with the tracker.
In a second step, the video camera (C) is calibrated with respect to the tracking system (K). This is achieved, as shown
in Figure 1, using a common reference marker (R) that is visible (only at the calibration step) to both cameras. A second
marker (M) is fixed to the tablet PC to allow for tracking it during the operation. The arrows show the transformations
between the different coordinate systems. For example, Tkr is the transformation from the reference R to the tracker K.
Following this chain of transformations, the transformation from the marker M to the video camera C can be calculated
as follows:
Tmc = Tcr .Tkr-1.Tkm
(1)
This one-step calibration is performed fully automatically. The resulting transformation remains valid as long as the
marker M does not move relative to the video camera. At operation time, the patient is fixed with respect to the reference marker, which, in turn, will no more be visible to the video camera C. Fortunately, the transformation Tcr can now
be computed using:
Tcr = Tmc .Tkm-1.Tkr
(2)
Since Tmc is constant from the previous step and Tkr is inherently fixed, Tcr is only dependent on Tkm, i.e. on the tracking
data of the marker M.
1
http://www.vision.caltech.edu/bouguetj/calib_doc/
126
Proceedings curac2010@MEDICA
To create correct correspondences between the patient (phantom model) and the scanned dataset, we adapt a pairedpoint rigid registration scheme. In essence, we choose a set of points in the patient coordinate system whose positions
can be precisely defined by the tracking system. A second set of corresponding points is interactively selected from the
scanned dataset (in our case, four points are selected from a scanned CT image). Absolute positions of these points are
calculated considering their voxel coordinates and image spacing. The two sets of points are finally registered utilizing
a least square fitting approach. Particularly, the rigid registration transformation Treg is composed as:
Treg = R*T
(3)
Here, T is a translation calculated through the shift of centroids of the two input point sets, and R is calculated using
singular value decomposition, as described in [1].
At rendering time, the frame captured by the video camera is first rendered as a background. Additional (virtual) objects
are then rendered/overlaid using OpenGL, where the MODELVIEW Matrix is computed as Tcr .Treg. Examples of objects that can be overlaid are 3D reconstructions of segmented structures like blood vessels and tomographical slices
from the anatomy dataset. For certain structures, e.g. tumors, an enhanced visualization of shape and depth information
can also be provided. This is achieved by extracting the planar contours of the tumor at successive depths perpendicularly to the viewing direction, as shown in Figure 2b. For this purpose, we calculate the intersection of the tumor surface
model with a cutting plane, while changing the depth of the plane. Depth information is conveyed via depth cueing, i.e.
the transparency of a contour is defined as a linear function of its depth so that deeper contours appear more transparent.
This contour visualization assists the surgeon to better determine the shape and extent of the tumor at different depth
levels.
To visualize tomographical slices, we adapt an optimized cross sectioning algorithm [5] to compute the slices at the desired position and orientation. This algorithm applies an incremental approach to compute, for each pixel location in the
cross section plane, the corresponding 3D location in the volumetric data, and evaluates the final pixel intensity using
tri-linear interpolation. The generated slice image is then used as texture for a rectangle, which is in turn rendered in the
real scene with the correct pose. Thereby, the spatial resolution (dimension and spacing) of the dataset and the cross
section are considered in order to exactly reflect the spatial proportion with the patient and environment, whose coordinate system is represented in millimeter.
(a)
(b)
Figure 2: Augmented views from the presented system: (a) A tumor is blended over the optical image; (b) Contours of
the tumor at successive depth levels (distance=2mm), deeper contours are more transparent.
Proceedings curac2010@MEDICA
127
Figure 3: GUI of the visualization module: The widget shows a tomographical slice that is computed perpendicular to
the viewing vector and superimposed on the operation field, together with an augmented display of the tumor contours.
3
Implementation and Results
The software components of our system have been implemented with C++ (Microsoft Visual Studio 2005). Rendering
was performed using OpenGL. The GUI, shown in Figure 3, was implemented using the Qt2 library. For our current
implementation of the prototype, we relied on marker-based tracking provided by the ARToolkit3, which allows for
multiple-marker tracking in real time. However, due to the inaccurate calibration provided by ARToolkit, we decided to
recalibrate camera parameters using MATLAB, as stated in the previous section. Consequently, the MATLAB based
calibration was much more accurate regarding 3D overlay and depth estimation processes.
To simulate an operation scenario, a phantom model of the head is CT-scanned and an imaginary tumor is manually
embedded in the CT dataset. The tumor is segmented from the dataset, partially based on Kitware’s Insight Segmentation and Registration Toolkit ITK4, and a 3D model is reconstructed using the Marching Cubes algorithm. Afterwards,
the video camera is calibrated with respect to the tracker (extrinsic calibration). After this one-step calibration, the reference marker R must stay fixed with respect to the tracker camera. Also, the tracking marker M must stay fixed with
respect to the video camera. The phantom model is then registered to the scanned data, using a small set of correspondence point pairs, including the corners of the reference R. Finally, augmentation into the video stream can start.
Figure 2 shows some snapshots of our prototype for enhanced visualization in a brain tumor surgery scenario. In Figure
2a, an augmented view of a malignant tumor (green) is presented. In Figure 2b, boundaries of the tumor at successive
depth planes (distance=2mm) are displayed, where transparency is defined as a function of depth. Figure 3 depicts the
2
http://qt.nokia.com/
http://www.hitl.washington.edu/artoolkit/
4
http://www.itk.org/ 3
128
Proceedings curac2010@MEDICA
GUI of the whole visualization module, showing an augmented view of the tumor contours (yellow) and an interactively selected tomographical slice from the CT dataset, blended over the optical image. In this context, the cross sectioning
algorithm allows for a computation and rendering rate of about 30 fps.
4
Discussion
We introduced a prototype for enhancing intraoperative visualization within the operating microscope, by augmenting
the video stream with relevant operational information. This prototype is a PC / tablet PC based simulation of the aimed
scenario using a surgery microscope. The essential components of the system (tracking, calibration, registration, and
rendering) have been described. The objects of interest are extracted from registered pre- and intraoperative patient data. This results in a better localization of targeted lesions and the protection of proximate risk structures, like blood vessels or certain functional units. From surgical point of view, the system offers a good usability and added value to the
procedure. Our next step is the transfer of the approach into a real surgery suite. Thereby, the surgical microscope will
replace our tablet PC based optical/visualization device, and the camera tracking will be performed by an infrared tracking system. Although the system initially was aimed for neurosurgery applications like brain tumor and hypophysis surgeries, the concept can be transferred to several other surgery procedures.
Acknowledgment
This work is funded by the German Ministry of Education and Science (BMBF) within the ViERforES project
(no. 01IM08003C).
5
References
[1] K. Arun, T. Huang, and S. Blostein, Least-Squares Fitting of Two 3D Point Sets, IEEE Transactions on Pattern
Analysis and Machine Intelligence, 9(5):698–700, 1987.
[2] W. Birkfellner, M. Figl, K. Huber, F. Watzinger, F. Wanschitz, J. Hummel, R. Hanel, W. Greimel, P. Homolka, R.
Ewers, and H. Bergmann. A Head-Mounted Operating Binocular for Augmented Reality Visualization in Medicine-Design and Initial Evaluation. IEEE Transactions on Medical Imaging, 21(8):991–997, 2002.
[3] P. Edwards, A. King, C. Maurer, D. de Cunha, D. Hawkes D. Hill, R. Gaston, M. Fenlon, S. Chandra, A. Strong, C.
Chandler, am M. Gleeson, Design and Evaluation of a System for Microscope-Assisted Guided Interventions
(MAGI). IEEE Transactions on Medical Imaging, 19(11):1082–1093, 2000.
[4] R. Lapeer, M. Chen, G. Gonzalez, A. Linney, and G. Alusi. Image-Enhanced Surgical Navigation for Endoscopic
Sinus Surgery: Evaluating Calibration, Registration and Tracking. The International Journal of Medical Robotics
and Computer Assisted Surgery, 4(1):32–45, 2008.
[5] Z. Salah, B. Preim, and G. Rose. An Approach for Enhanced Slice Visualization utilizing Augmented Reality: Algorithms and Applications, Proc. of PICCIT, 2010.
[6] R. Shahidi, M. Bax, C. Maurer, J. Johnson, E. Wilkinson, B. Wang, J. West, M. Citardi, K. Manwaring, and R.
Khadem. Implementation, Calibration and Accuracy Testing of an Image-Enhanced Endoscopy System. IEEE
Transactions on Medical Imaging, 21(12):1524–1535, 2002.
[7] F. Wacker, S. Vogt, A. Khamene, J. Jesberger, S. Nour, D. Elgort, F. Sauer, J. Duerk, and J. Lewin. An Augmented
Reality System for MR Image-guided Needle Biopsy: Initial Results in a Swine Model. Radiology, 238(2):497–
504, 2006.
Proceedings curac2010@MEDICA
129